Upgrading requires a full cluster restart. Note, once upgraded, there is no option to downgrade to previous version as the index files format will be also upgraded (new lucene version).

Breaking changes:

• Query Parser Module Refactor (#994)
• ‘,’ can be used in type name (#996)
• Analysis: Phonetic filter replace flag is inverted, fix (changes default behavior now, to follow docs) (#1127)
• Field Cache: Change default type to resident from soft (#958)
• Filter Cache: Introduce new node level filter cache and make it default (#959)
• Node Stats: Remove low level transport stats from response (#979)
• Thrift: Upgrade to thrift 0.6.1 (#986)

New features:

• Upgrade to Lucene 3.2 (#997)
• New default merge policy – tiered (#998)
• Indices exists API (#1022)
• Realtime GET (#1060)
• Multi GET API (#1068)
• Upgrade to Lucene 3.3 (#1083)
• Nested Object/Docs Mapping and Searching (#1095)
• Nested Objects Facets Support (#1098)
• MappingCharFilter (#1106)
• Allow to disable automatic date detection (#1051)
• Aliases:
  • add an ability to specify filters on aliases (#971)
  • allow to specify routing value for an alias (#1025)
• Get API: Make type optional (#1061)
• Indices Segments API: Internal segments info of shard level Lucene indices (#1064)
• Mapping: Document’s Field level boosting (#920)
• Nodes APIs: All node APIs to allow to match on nodes based on addresses, names, and attributes (#1125)
• Plugins: Allow plugins to serve a _site, automatically download github plugins (#978)
• Query DSL: nested filter support (#1102)
• Search: Optimize (perf) execution of global facets (#889)
• Zen Discovery: Add minimum_master_nodes setting helping with split brains (#1079)

Enhancements:

• (Small) optimization to has_child filter (#880)
• Week Interval on Date Histogram Facet (#904)
• Simplified Disable Flush operation (#906)
• Upgrade to jackson 1.8.1 (#941)
• Request for more values identified as boolean for the boolean field type – no (#984)
• Analyzers for FLT / MLT (#1031)
• Add ec2 specific network hosts (#1037)
• MappingCharFilter (#1107)
• PatternReplaceFilter (#1108)
• Closed indices should not cause block failures on some APIs what executing against _all indices (#1010)
• Add ‘other_terms’ option for terms facet (#1029)
• Allow to configure node name using node.name (on top of current name) (#1126)
• Aliases: Validate alias filter before adding it (#993)
• Analysis:
  • Expose light and minimal language token filters (#908)
  • Stop words config to allow for automatic lang expansion (#909)
  • Word Delimiter Token Filter (#918)
  • All analysis components that accept stopwords to allow to load stopwords from a file (#919)
  • Synonym Token Filter (#900)
  • Pattern Tokenizer (#928)
  • Add elision token filter (#930)
  • Phonetic filter to support cologne encoder (#942)
  • Improve custom analyzer construction time (#989)
  • Add reverse and skip to path_hierarchy tokenizer (#999)
  • expose Lucene KeywordMarkerFilter (#1052)
  • expose Lucene StemmerOverrideFilter (#1053)
  • backport KStem token filter from Lucene (#1041)
• Create Index API: Better logic to wait for ack for indices being created on nodes (#885)
• Field Cache: Allow to set index.cache.field.expire when using resident type (#957)
• Get Mapping: Better error response when asking for specific index type (#991)
• Mapping: non-string type field level boosting (#1097)
• Node Info / Stats: Add max_file_descriptors (info) and open_file_descriptors (stats) (#1113)
• Node Stats: Add number of server open channels for transport and http (#1115)
• Numeric Types: By default, omit_norms (boosting) is set to true, automatically set it to false if explicit field mapping boosting is provided (#1021)
• Phonetic Filter : Double Metaphone, partial implementation (not using the secondary code) (#924)
• Query DSL:
  • Allow to set boost on has_child query (#884)
  • Fuzzy Query – add max_expansions (#905)
  • Fuzzy query support for numeric / date types (#907)
  • Text Queries (boolean, phrase, and phrase_prefix) (#917)
  • query_string – Expose QueryParser#setAutoGeneratePhraseQueries (#964)
  • Ids Filter / Query – allow to execute it with no type defined / several types (#969)
  • Add limit filter (#976)
  • Simplify not filter to accept the filter directly, without a wrapping filter element (#992)
• RabbitMQ River: Upgrade to rabbitmq driver 2.5.1 (#1112)
• Scripting: Optimize to native script execution when using just doc.score (#952)
• Search: When a search request failed completely (all shards fail) return a proper HTTP status code (#1035)
• Shared Gateway: Allow to dynamically update the snapshot_interval using update settings (#954)
• TCP: Change the default connect timeout to 2s (#1104)
• Transport: Improve concurrency when connecting to several nodes (#1007)
• Unicast discovery: Improve initial ping connection to nodes to be more lightweight (#1028)
• Update Settings:
  • Properly ignore settings that are not allowed to be updated dynamically (#985)
  • Allow to change non dynamic settings on a closed index (#1048)

Bug fixes:

• include_in_all not overridable in objects (#883)
• Sort missing wrongly sorts negative values (#899)
• Possible failure to start a river after cluster restart (#902)
• Failure when applying failed shards during local gateway allocation (#903)
• Possible failure when using TransportClient (with sniffing) (#923)
• Deleting and recreating a new index with dynamic mapping can cause type failures (#931)
• wrong type returned by date_histogram (#932)
• IBM J9 failure to serialize nodes info (#950)
• NPE when using “not” filter (#953)
• fix plugin.bat classpath (#1002)
• Merging complex objects / types (geo, objects) can cause failure to lookup field names (without type prefix) (#1016)
• Numeric values always omit norms (boosting), regardless of the configuration set (#1020)
• River might not start properly after cluster shutdown (timing) (#1038)
• Misbehaving “missing” field in facet results (#1070)
• Without pre-creating index, heavy write rates OOM with too many threads (#1081)
• java.lang.ArrayIndexOutOfBoundsException when indexing a doc (#1094)
• Uneven distribution of search requests across shards (#1103)
• Native (java) process memory leak (#1118)
• Index / Delete API when causing auto index creation can cause buffer data overrun (#1122)
• Analysis: Regression (0.16.1), Camel cased filters / tokenizers failed to load in custom analyzer (#937)
• And/Or Filter: Possible failure when inner filter does not match any docs (#892)
• Delete API: Using refresh parameter might not refresh delete operation on replica (#1047)
• Filter Cache: soft filter cache can cause bad memory behavior (#946)
• Mapping: dynamic templates of object type do not initialize some mapping data structures (like analyzers lookup) (#929)
• Mapping – Date Format: 3 or more custom date formats fail using ‘||’ delimiter (#977)
• Parent Child: complex parent child mapping (>2 hierarchy) might fail (#1039)
• Put Mapping: Fix rare case where the response will timeout (ack=false) even though it has been applied (#990)
• Query DSL: Span Term Query wrongly parses when boost is provided, causing using it in span or to fail (#1032)
• Scripting – mvel: Fix thread safety misuse when compiling scripts (#886)
• Search:
  • Empty facets element causes search failures (#938)
  • When all shards fail on second / third phase, make sure to return a response with proper shard failures (#1033)
• Search API: filter element should not be applied to global facets (#888)
• Transport Client: Adding more nodes causes more scheduled reconnect tasks (#1062)
• auto_expand_replicas: [0-all] can cause data loss when nodes are removed (#934)